Accelerated Natural Language Processing Lecture 2 Morphology - - PowerPoint PPT Presentation

Accelerated Natural Language Processing Lecture 2 Morphology

Sharon Goldwater (based on slides by Philipp Koehn) 17 September 2019

Sharon Goldwater ANLP Lecture 2 17 September 2019

Two plots from last time

Sharon Goldwater ANLP Lecture 2 1

How Many Different Words?

10,000 sentences from the Europarl corpus

Language Different words English 16k French 22k Dutch 24k Italian 25k Portuguese 26k Spanish 26k Danish 29k Swedish 30k German 32k Greek 33k Finnish 55k

Why the difference? Morphology.

Sharon Goldwater ANLP Lecture 2 2

Today’s Lecture

• What is morphology, how does it differ across languages, and why

does it matter for NLP?

• What’s the difference between a stem, lemma, and affix?
• What are the characteristics of derivational and inflectional

morphology?

• What is an FSM, and what is the relationship between FSMs and

regular languages?

Sharon Goldwater ANLP Lecture 2 3

Interlude/reminder: types and tokens

The word word is ambiguous.

• Word type:

“10k sentences from English Europarl have 16k different words” (unique strings, lexical items)

• Word token:

“English Europarl has 54m words” (possibly repeated instances) a cat and a brown dog chased a black dog: 10 tokens, 7 types.

Sharon Goldwater ANLP Lecture 2 4

What is morphology?

The study of wordforms and word formation.

• Structured relationships between words:

play, played, replay, player played, walked, jumped

• Units
• f

meaning (morphemes) and their

• rdering

(morphotactics): de+salin+ate+ion but not ate+salin+ion+de

Sharon Goldwater ANLP Lecture 2 5

Why does morphology matter?

• Information retrieval: return pages with related forms.
• Language modelling: make predictions about unseen words
• Machine

translation and language understanding: signals differences in meaning (might be expressed using word order in other languages).

Sharon Goldwater ANLP Lecture 2 6

Why does morphology matter?

Example (Russian): zhenshina devochke dala knigu woman+NOM girl+DAT gave book+ACC ‘the woman gave the girl a book’ vs. zhenshine devochka dala knigu woman+DAT girl+NOM gave book+ACC ‘the girl gave the woman a book’ A noun’s case marking (a kind of morphology) indicates its role in the sentence, where English uses word order and prepositions.

Sharon Goldwater ANLP Lecture 2 7

Morphemes: Stems and Affixes

• Two types of morphemes

– stems: small, cat, walk – affixes: +ed, un+

• Four types of affixes

– suffix – prefix – infix – circumfix

Sharon Goldwater ANLP Lecture 2 8

Stems vs. Lemmas

• Lemma: the canonical form or dictionary form of a set of words

– fly, flies, flew and flying all have the lemma fly. – walk, walks, walked and walking all have the lemma walk. – walker, walkers have the lemma walker.

Sharon Goldwater ANLP Lecture 2 9

Stems vs. Lemmas

• Lemma: the canonical form or dictionary form of a set of words

– fly, flies, flew and flying all have the lemma fly. – walk, walks, walked and walking all have the lemma walk. – walker, walkers have the lemma walker.

• Stem: definitions can vary, but often: the part of the word that

is common to all its variants – stem of produce, production is produc. – stem of walk, walks, walked, walking, walker, walkers is walk. – Do fly, flies, flew, flying have a common stem fl? Or maybe only fly and flying share a stem: fly. Decision may depend on application.

Sharon Goldwater ANLP Lecture 2 10

Suffix

• Plural of nouns

cat+s

• Comparative and superlative of adjectives

small+er

• Formation of adverbs

great+ly

• Verb tenses

walk+ed

• All inflectional morphology in English uses suffixes

Sharon Goldwater ANLP Lecture 2 11

Prefix

• In English: these typically change the meaning
• Adjectives

un+friendly dis+interested

• Verbs

re+consider

• Some language use prefixing much more widely

Sharon Goldwater ANLP Lecture 2 12

Other types of morphology

Mainly in non-English languages; check textbook or online.

• Infixes
• Circumfixes
• Reduplication
• Root and pattern

Sharon Goldwater ANLP Lecture 2 13

Not that easy...

• Affixes are not always simply attached
• In writing, some letters may be changed/added/removed

– walk+ed – frame+d – emit+ted – carr(–y)+ied

• In speaking, some sounds may be changed/added/removed

– Compare the final sound: cats [s] vs dogs [z] vs foxes [@z]

Sharon Goldwater ANLP Lecture 2 14

Irregular Forms

• Some words have irregular forms:

– is, was, been – eat, ate, eaten – go, went, gone

• Irregular forms tend to be the most frequent (and vice versa)

Sharon Goldwater ANLP Lecture 2 15

Inflectional Morphology

• In English, we inflect

– nouns for count (plural: +s) and for possessive case (+’s) – verbs for tense (+ed, +ing) and a special 3rd person singular present form (+s) – adjectives in comparative (+er) and superlative (+est) forms.

• In German, we inflect

– nouns for count and case – verbs for tense, person, and count – adjectives for count, case, gender, and definiteness – determiners for count, case and gender

Sharon Goldwater ANLP Lecture 2 16

Forms of the German the

Case Singular Plural male fem. n. male fem. n. nominative (subject) der die das die die die genitive (possessive) des der des der der der dative (indirect object) dem der dem den den den accusative (direct object) den die das die die die Phrase/role: [The A]/s put [the B]/o [of the C]/p [on the D]/io Not only many different forms, but each form is highly ambiguous.

Sharon Goldwater ANLP Lecture 2 17

Inflectional vs. Derivational Morphology

• Inflectional morphology typically

– does not change basic meaning or part of speech – expresses grammatical features or relations between words – applies to all words of the same part of speech

• Derivational morphology

– may change the part of speech or meaning of a word – is not driven by syntactic relations outside the word – may be “picky”: drama+(t)ize but not traged(-y)+ize – applies closer to the stem; whereas inflection occurs at word edges: govern+ment+s, centr+al+ize+d

Sharon Goldwater ANLP Lecture 2 18

Derivational Morphology

• Changing the part of speech, e.g. noun to verb

word → wordify

• Is it a real word?
• Consulting Google (a few years ago):

– 8,840 hits: e.g., wordify mugs, tshirts and magnets

• Google now returns over 75k hits. (Why?)

Sharon Goldwater ANLP Lecture 2 19

Derivational Morphology

• Changing the verb back to a noun

wordify → wordification (8k hits on Google)

• A person/thing who engages in wordification

wordification → wordificator (was 8 hits, now 21k: another app!)

• A person/thing who wordifies

wordify → wordifier (1500 hits on Google)

• What is the difference between a wordifier and a wordificator?

Sharon Goldwater ANLP Lecture 2 20

Derivational Morphology

• Turning wordification into a ideology:

wordification → wordificationism (was just 1 hit:) I think you’re confusing the term “Democracy” with “Capitalism”; I think you mean “Has Capitalism failed”?

• No. It hasn’t.

I agree, Hambone; I’m just trying to correct the wordificationism. Where in the world did you get the word “wordificationism”? Not in the Merriam-Webster dictionary, not in the Thesaurus...

Sharon Goldwater ANLP Lecture 2 21

Derivational Morphology

• An adherent of wordificationism

wordificationism → wordificationist

• Used to have 0 hits on Google, now you get these slides!
• We created a new word!

Sharon Goldwater ANLP Lecture 2 22

Compounds

• Creating new words by merging multiple words
• (Somewhat) rare in English

home work → homework web site → website

• More common in other languages (like German)

Sharon Goldwater ANLP Lecture 2 23

Acronyms/Initialisms

• Wikileaks / Guardian, document 2007-081-100110-0444:

OGA

• perating

in TF Catamount sector moved into Malekshay for operation. LN Shum Khan ran at the sight of the approaching CFA’s. CF utilized the escalation of force doctrine and shouted to stop, fired warning shots and then fired to wound. The LN was hit in the ankle and treated by Element medics on scene. It was determined through discussions with local Elders that the man was a deaf mute that was nervous of the CF operation. Solatia was made in the form of supplies and the Element mission progressed

Sharon Goldwater ANLP Lecture 2 24

Morphology differs across languages

• Usually a trade-off between morphology and syntax (word order)

– Some languages have no verb tenses → use explicit time references (yesterday) – Case inflection determines roles of noun phrase → use fixed word order instead → use prepositional phrases instead of cased noun phrases

• Examples from the World Atlas of Language Structures (wals.info)

– prefixes vs. suffixes – cases (zero to more than ten) – past tense remoteness distinctions

Sharon Goldwater ANLP Lecture 2 25

Sharon Goldwater ANLP Lecture 2 26

Sharon Goldwater ANLP Lecture 2 27

Sharon Goldwater ANLP Lecture 2 28

So...

How to deal with all this computationally? What do we even want to be able to do?

Sharon Goldwater ANLP Lecture 2 29

Tasks

• Recognition

– given: wordform (string of characters) – wanted: yes/no decision if it is in the language

• Generation

– given: lemma and morphological properties – wanted: correctly inflected wordform

• Analysis

– given: wordform – wanted: lemma and morphological properties

Sharon Goldwater ANLP Lecture 2 30

Word Lists

• Simple Solution

– create a list of all wordforms and their morphological properties – solve tasks by checking against list

• But...

– list can become very long – list fails to generalize for productive morphology

• Instead: use finite state machines

(also called finite state automatons)

Sharon Goldwater ANLP Lecture 2 31

Finite State Machines: States

places we may find ourselves in

Sharon Goldwater ANLP Lecture 2 32

Finite State Machines: Transitions

moving between the states

Sharon Goldwater ANLP Lecture 2 33

Finite State Machines: Emissions

a b a b b c c a c a c b

emissions: letters produced at each transition

Sharon Goldwater ANLP Lecture 2 34

Finite State Machines: Start and End

START END a b a b b c c a c a c b

begin at start state, finish at end state

Sharon Goldwater ANLP Lecture 2 35

The language of an FSM

Every FSM defines a formal language:

• The set of strings that can be generated by moving from start to

end states, emitting symbols on each transition.

• Equivalently, the set of strings that can be recognized by

matching input characters to emission symbols. The language of an FSM may be finite or infinite.

Sharon Goldwater ANLP Lecture 2 36

FSM with Finite Language

START END a b a b b c c a c a c b

generated language: { acac, acbc, aacc, aabb, bacc, babb }

Sharon Goldwater ANLP Lecture 2 37

FSM with Infinite Language

START END a b a b b c c a c a c b b

generated language: { acac, acbc, aacc, aabb, bacc, babb, bbacc, bbabb, bbbacc, bbbabb, bbbbacc, bbbbabb, ... }

Sharon Goldwater ANLP Lecture 2 38

Regular Languages

• Languages produced by FSMs are called regular languages
• Many convenient properties (e.g., straightforward to determine if

a word is in the language)

• Not all languages are regular

example: anbn = { ab, aabb, aaabbb, aaaabbbb, ... } (would require an infinite number of states)

Sharon Goldwater ANLP Lecture 2 39

Regular Expressions

• Reg. languages can also be described with regular expressions.
• Every RegEx is equivalent to some FSM (and vice versa).

Example: ac(ac|bc) | aa(cc|bb) | bb∗a(bb|cc) where ‘|’ means “or” and ‘x∗’ means “zero or more x’s”.

• RegExs are common in programming to describe sets of strings.

– ls *.jpg – if ($word =~ /^[A-Z].*/) { $name = 1; } – if ($name =~ /[WB]ill/) { print "Will or Bill"; }

Sharon Goldwater ANLP Lecture 2 40

Chomsky Hierarchy

• Chomsky discussed four major classes of formal languages
• 3. regular (generated by finite state machines, usually assumed

sufficient to describe phonology and morphology)

• 2. context-free (will be covered in later lectures on syntax)
• 1. context-sensitive (possibly needed for some natural language

phenomena)

• 0. recursively enumerable (anything a computer program can

produce)

• (There are also many classes of “sub-regular” languages.)

Sharon Goldwater ANLP Lecture 2 41

Chomsky Hierarchy

• Language classes further down the list are increasingly complex

– can describe more languages – but languages in the class are more difficult to compute for instance: for a type-0 language it is not generally possible to determine if a specified word can be generated by the language

• Linguists argue about which (if any) of these classes natural

languages belong to, but most phenomena of interest can be described by context-free languages.

Sharon Goldwater ANLP Lecture 2 42

Questions for review

• What is morphology, how does it differ across languages, and why

does it matter for NLP?

• What’s the difference between a stem, lemma, and affix?
• What are the characteristics of derivational and inflectional

morphology?

• What is an FSM, and what is the relationship between FSMs and

regular languages?

• (To be answered next time: how do we use FSMs for morphology?)

Sharon Goldwater ANLP Lecture 2 43

Exercises

• 1. Look at the FSM on slide 38, where there is a state that has a

self-loop labelled ’b’. Suppose we added another self-loop to the same state, labelled ’c’. Which of the following strings is NOT accepted by the new FSM? acac aacc bacc bbacc bcacc bcbabb bacbb bccabb

• 2. What is the lemma of each of the following words?

How many affixes does each word have, and are they derivational

• r inflectional?

located dreamy stole standardizes

Sharon Goldwater ANLP Lecture 2 44

Reminders

• 1. Labs this week:
• Check Learn to see which lab to go to and for additional prep

instructions.

• When you arrive in lab, sit down and work with a partner!

Discuss and help each other. Pass the keyboard back and forth.

• 2. Tutorials next week (starting Tuesday):
• You’ll be automatically enrolled in one when you register, so

try to do that by the end of this week.

• Also, start going through probability tutorial (linked from week

2 Reading).

Sharon Goldwater ANLP Lecture 2 45